학술논문

A low-power partitioning methodology by maximizing sleep time and minimizing cut nets
Document Type
Conference
Source
Fifth International Workshop on System-on-Chip for Real-Time Applications (IWSOC'05) System-on-Chip for Real-Time Applications System-on-Chip for Real-Time Applications, 2005. Proceedings. Fifth International Workshop on. :368-371 2005
Subject
Computing and Processing
Components, Circuits, Devices and Systems
Sleep
Circuits
Partitioning algorithms
Iterative algorithms
Databases
Data engineering
Application software
Minimization
Linear programming
Partitioning
Subthreshold Leakage Power
Genetic Algorithm
Sleep Time
Geometric Iterative
Improvement
Segmented Trees
Language
Abstract
The rising objective in VLSI design is to minimize the average power consumption. Sleep time maximization along with minimization of cut nets are explored as ways to decrease and minimize the power consumption. The major motivation is to deactivate parts of a circuit when they are idle, while simultaneously keeping the cut nets as low as possible. This dual objective problem is separately formulated as two single objectives and then combined into one normalized objective function. The joint problem is shown to be NP-hard, hence heuristic approaches were introduced. A modified version of the genetic algorithm is presented along side with an efficient implementation of a geometric iterative improvement technique using segmented trees. Results are presented for three hypothetical test cases and the results demonstrate more than 40% improvement.